The plan for my next week

[Les]

Forget "die sim"
We have still not solved a MCW6000 on a simple uniform flux "die sim".(link)
Have little hope of progressing with more complex until solved.
Maybe with input from experienced modellers progress will be made - but no sign of such input.

Would still be happy to play with any numbers coming from an Incoherent experimental set-up

[BillA]

Les
the 6000 is probably the worst wb you could select, much more complex than it seems (off ctr, recirculation, bypass; lots of fluid manipulation)
why not start with the SlitEdge then WW (SE w/ctr inlet), much saner flow regimes

[Les]

The only wb Inco has.
He's the boss
Would also be nice to use a wb where dimensions are known and not hearsay,but...........

[BillA]

oh crap
I'm guessing to that he wants to pursue modeling ?
then I would suggest something that he wants to test, a wb having a side to side flow; then impingment
the 6000 is a hybrid, totally useless for parametric modeling

[Les]

Am unsure.
He's being making tentative exploratory ventures into the Apogee thread .
However I think he is still feeling his way and only a model dilettante, better than my 13 flip beermat but sill a dabbler .
Being a one-man band it means he has to commit to both interpretation and implementation.
= think you know that game.
Dunno

Edit "a wb having a side to side flow; then impingment
.........."
The way I hoped we were going 3 years ago, but the main goal then became to get the best wb and ignore understanding

[Marci]

The only wb Inco has.
He's the boss

Can supply with TDX, WW, RBX and a few others if needs be - got plenty of `em...
Lemme know...

[Les]

He is his own boss.
I am only a guest in his threads.
He chooses his experiments to suit his interests and his capabilities.
However,personally, I would have no interest in WW, RBX or TDX.
WW is a beast to establish a Primary Convection Area, RBX and TDX are convoluted affairs which have never modelled.
Maybe the Slitedge would be interesting if wished to start with clean sheet and remodel from scratch,but not a good idea whilst I still use beermat modelling. Would be quicker and better done by 21st century modellers;the inactive Groth and gmat spring to mind. Sadly no-one since has volunteered any finite element analysis(or whatever these techniques are called)at the sharp-end. Data sits in In the Flux-bloc thread( (link again) awaiting analysis.Plenty talk but no action.

Hate these wordy posts

[Les]

Withdrawn Attachment - Errors.Done Something Stupid and Can't Find *

First guess at AMD 3000+ model.
Used experimental Parabolic Flux model which for the moment has forced me onto Circular dies
Modelled for TIM 1 of 0.06c/w per cm^2 and the old Intel(ish) 0.25c/w per cm^2 TIM, linked by Marci link
Guess AMD probably use a TIM intermediate between 0.06 and 0.25.
Tendency to degrade as in article? Easier to use a bigger than 101mm^2 die ?
More to be added to attachments: 1mm HotSpots on far-surface of 0.5mm thick 148w/mK slice of Silicon
w/o IHS model

* Edit:
The root of the prob which caused withdrawing of attachment for my model AMD3000+ :-
Get different solutions when considering "bp +IHS" as one entity and calculating separately .
e.g Apogee(34diam x3)+TIM(0.06c/w*cm^2) + IHS(34diamx1mm) on 11.34 diam Source.
At h(eff)=43788 and IsoFlux
Consider as continuous 4mm or continuous (3+1)mm then adding TIM get: R(water -> far IHS surface) = 0.112777+0.006607=0.11938
Considered as (3mm with h=h(eff)) + (1mm with h=1/(Rbp+1/h(eff)*A+Rtim)*A get: R(water -> far IHS surface) = 0.156659



Still hoping for comments on "Anyone know undercup bp thickness of the Storm G4 ?
Have upped from 0.7625mm to 1.2625mm following Cathar's "Storm's bp is substantially thicker than 1mm" (link to page).
Is this right?"

[BillA]

gee Les, it is a shame you've got so much in the 6000 - I had no idea you were treating it as other than a black box (did I not suggest such ?, shame on me if not)
must admit to amazement at those other wbs being considered 'convoluted' as compared to the 6000 (sure not in my mind, or by the experimentation done)
only active CFDer I'm aware of is Stephen, who is muted and whose first efforts got caught in Swiftech's pr fiasco

I am at a loss to imagine what on what basis one would model a 'hot spot', more smoke and mirrors than I can count

[Les]

Black box is useless(IMO) when comes to the nitty-gritty of Die-simulators.
Incoherent only had MCW6000 - hence the only Virgin WB data(No TIM, no unknown offset)
Need a "Primary Convection Area" and bp thickness(t) before can analyze* for h(eff).
For Storm ,say, is easy to get a 1st guess by taking cupped area +smidgen.
Usually have to model from beginning to get Apca(even Storm have to estimate profile outside cupped area - I ignore>0.5mm)
Apca and h(eff) being only wb and flow-rate dependant can then begin.

* Need Apca and t to get h(eff) from R(virgin)=1/Apca*h(eff) + (Rcu.1-d) + (Rcu.spreading).
Also from begiining need a 1st guess to interpret Sensors T data - see Inco here for example

Edit: Added "*" expansion

[BillA]

I would that others paid as much attention to Inco as you
as to the comment about the sink affecting the source, is that not the purpose ?
it is precisely your PCA (Apca ?) that is so difficult with the 6000; semi-impingement, flow bypass, huge swept area, thick bp
anything you select is no more than that, a selection
to have a realistic basis some parametric modeling is needed to establish the sensitivity to source size (uniform flux IS assumed)
you are doing it backwards ? (do I understand ?)

[Incoherent]

Originally Posted by BillA

to have a realistic basis some parametric modeling is needed to establish the sensitivity to source size (uniform flux IS assumed)
you are doing it backwards ? (do I understand ?)

The problem is the temperature sensor. It's reading is dependent on it's position and the flow of heat around it. The flow of heat around it is dependent on block geometry. So, to make sense of the reading the block geometry needs to be known, AND modelled accurately to back calculate the temperature at location as if the sensor had not been there . THEN we have established the waterblock performance.


The Source size is an interesting model in itself. This I will start with I think. With no sensors and with sensors in various configurations.

FEMLAB session tonight.

[BillA]

do you mean "the block geometry needs to be known" in the sense of Les' Apca ?
would this not be the 2 (source and sink) being more/less (+ and -) similar ?

[Les]

Originally Posted by BillA

I would that others paid as much attention to Inco as you
as to the comment about the sink affecting the source, is that not the purpose ?
it is precisely your PCA (Apca ?) that is so difficult with the 6000; semi-impingement, flow bypass, huge swept area, thick bp
anything you select is no more than that, a selection
to have a realistic basis some parametric modeling is needed to establish the sensitivity to source size (uniform flux IS assumed)
you are doing it backwards ? (do I understand ?)

"the sink affecting the source, is that not the purpose ?"
Yes but was referring to sensors not source

"it is precisely your PCA (Apca ?) that is so difficult with the 6000; semi-impingement, flow bypass, huge swept area, thick bp anything you select is no more than that, a selection "
Yes it is a selection, then do beermat iteration with old Swiftech data - pick-and mix.
The plateau is a liberal interpretation of Suresh Garimella's work .The extent and height of plateau and height of descent-from are Flomerics determined(should be fudge factors but..)
Apogee is relatively easy in that plateau taken to extend beyond pin area.
WW and MCW are beasts.
Similarly MP-05-SP the plateau controlled by outer jets.

"(uniform flux IS assumed)"
Yes in die-simulators. Trying parabolic for CPUs

"you are doing it back wards ? (do I understand ?)"
Yes, in so far as am still using modelled parameters in all model extensions.
This seems reasonable with fair agreement between modelled and half analyzed flux-bloc-MCW6000 values of Rvirgin
When and if there are some experimental values for h(eff) will incorporate them.
Think you understand

Edit1:

Originally Posted by Incoherent

FEMLAB session tonight.

Best of luck..
My first attempts at Strand 7 only produced a pretty picture of a beam.
However have been forced into Excel formula - Parabolic W and Bill's radiators
Edit2: Added link to Garimella's work

[BillA]

do you 'see' (presume ?) the same/corresponding temp profile ?
not opining, asking

[Incoherent]

Originally Posted by BillA

do you mean "the block geometry needs to be known" in the sense of Les' Apca ?
would this not be the 2 (source and sink) being more/less (+ and -) similar ?

I hope that a "PCA" is enough. I am not yet at the level where I can model actual geometry and derive a local convection coefficient from the flow pattern.
Lets see what FEMLAB says about this then go from there.

<Threadjack. >

Some parameters for modelling wb + TIM + diesim or CPU
Suggestions.
I will model a 1/4 waterblock.
All TIMs: k=11.5 w/m°C, L=0.1mm. This gives about 0.0604 °C/W for a 12x12mm and is an easy unit for L. <0.1 mm gets too complex to model efficiently.

Uniform flux, CPU 1mm thick, Die 10mm thick heatsource 100W heat flux=100/A W/m^2
Die size 1 10x10mm
Die size 2 12x12mm
Die size 3 14x14mm
Die size 4 30x30mm

BP thickness

BP 1 (storm) 1.2mm
BP 3 ... 2.0mm?
BP 4 ... 3.0mm??
BP 2 (MCW?) 5.0mm


Les I need a function to describe "h(eff)" relative flowrate and one constant which is the block figure of merit. This is to help me quickly change the modelled scenario. What do you think of h=kx^0.8? x being flowrate, k being WB FOM. These are only parameters for the model. h(eff) will be the same for a given flowrate on different BP thicknesses.

Maybe 4 different FOMs giving h(eff) as shown in attached chart

The permutations here are... lots.

Speak now before I get too far or forever hold your peace.

Edits: (there might be a few of these in the form of suggestions required/requested)

Edit1: Please suggest density and heat capacity for TIM







.

[BillA]

make Tim1 75% of Tim2 if modeling for an Intel IHS w/solder joint (it is lower, this is a guess)

are you not going to wind up with an 'effective' thickness just as for area ?

[Les]

Originally Posted by BillA

do you 'see' (presume ?) the same/corresponding temp profile ?
not opining, asking

At the far-surface of the 0.5mm(assumed) Si die, I think yes.
Elsewhere would expect a modified profile.
It is one for FAE crowd. (Apologies if I've passed to you Johnathon. I don't FAE from FSEN. CSF or whatever)
I want an easier question

[Les]

"Les I need a function to describe "h(eff)" relative flowrate and one constant which is the block figure of merit. This is to help me quickly change the modelled scenario. What do you think of h=kx^0.8? x being flowrate, k being WB FOM. These are only parameters for the model. h(eff) will be the same for a given flowrate on different BP thicknesses."

Give me 1/2hr.

[Incoherent]

Originally Posted by BillA

make Tim1 75% of Tim2 if modeling for an Intel IHS w/solder joint (it is lower, this is a guess)

are you not going to wind up with an 'effective' thickness just as for area ?

Yes and no.

Baseplate thickness is distance from base to lowest part of structure. All copper above this is part of h(eff).
HOWEVER. Of course the temperature profile will not be even 1µm below this "slice", it will be affected by the geometry above, so in that respect the model is flawed. The h(eff) approach counts on uniformity here. Hopefully it spreads enough. Model mk II will look at this.

Damn, need a model to verify the model :/

One step at a time. For now a uniform h over flat surface, next a uniform h over a structured surface, then, hopefully, local h, structured surface. That is tough, but I hope doable when I am further up the learning curve.

Will do with the soldered TIM. It's Indium? need a k, Cp, density

[Les]

Not sure but
Do you want areas?

[Incoherent]

Originally Posted by Les

Not sure but
Do you want areas?

I'll use these Les. Area? Yes please.
And your assumed BP thickness. (to 0.1mm, smaller is painful to construct)

This makes it interesting. I'd need to make different areas of different h's, concentric circles/squares?, I'll start with one "patch"

[Les]

MP-05-SP 14.4x14.4 x1 mm
Storm G4 21.5x21.5x1.3
WWLE 17x15x 0.8 mainly but..
Apogee 34x34x3
MCW600* 40x40x5

[BillA]

no info on solder used, pick a ~150°C indium alloy ? (no worse assumption than any other ?)
not too sure the furniture can be separated from the bp, I understand to what you are referring (I thimk, lol)
- better to determine h(eff) for the flat bp experimentally for thickness/area as req to validate the model, of course by flow 'regime'

[Incoherent]

Les.

Your h(eff) vs flow chart, specifically the MP-05-SP?

How did you arrive at these?

The model using these numbers is showing it to be astounding.

Anyway, here are the model results of the three I have done so far. It seems to be behaving in a sensible manner, responds to parameters as it should. May be GIGO but that's all for tonight.

Feeding the model parameters is very similar to actually testing waterblocks for real in that I am changing flowrate etc. Just a lot less messy.


Edit: Well that was amusing. I just figured out how to input flowrate parameters automatically, and solve the whole "vs flowrate" model all at once. Just performed a 2 minute calculation that I had spent the last 3 hours doing manually. lol. RTFM idiot